Multi-Objective Reinforcement Learning-Based Dependent Task Scheduling With Service Caching in Mobile Edge Computing

doi:10.1109/TCCN.2026.3657056

Multi-Objective Reinforcement Learning-Based Dependent Task Scheduling With Service Caching in Mobile Edge Computing

2026-01-23

0

PRE

AI

DOI：10.1109/TCCN.2026.3657056

Original

Original request for help

Save

Abstract

En 中文

This paper investigates the dependent task scheduling with service caching (DTSSC) in mobile edge computing (MEC) systems, where each task requires a specific service program for execution. The DTSSC problem is characterized by bi-objective optimization, minimizing the application delay and energy consumption of the mobile equipment, simultaneously. The conflict between the two objectives in the problem makes it quite challenging to address. Recently, some single-objective reinforcement learning (SORL) algorithms have been introduced to solve the DTSSC problem. Nevertheless, these SORLs adopt the weighted sum method to define the user utility, thus ignoring the conflict between objectives. Furthermore, in dynamic MEC scenarios, the weights (i.e., preference) assigned to each objective may vary over time, posing significant challenges for conventional SORLs. Although several multi-objective reinforcement learning (MORL) algorithms have been adopted to handle dynamic preferences, they only adapt to small-scale dynamic preferences and cannot generalize across all possible preferences. To solve these challenges, we first build a multi-objective Markov decision process model that has a vectorial reward mechanism. Each component of the reward and one of the two objectives is correlated. Then, we propose a new trajectory-based experience replay scheme to improve sample efficiency and reduce replay buffer bias, resulting in a modified MORL algorithm. The experimental results demonstrate that the proposed algorithm is more adaptive to dynamic preferences and strikes a better balance between objectives compared with several algorithms.

Keywords:

Dependent task scheduling

dynamic preferences

mobile edge computing

multi-objective reinforcement learning

service caching

Multi-Objective Reinforcement Learning-Based Dependent Task Scheduling With Service Caching in Mobile Edge Computing

Abstract

Journal

Organization